Image Inpainting Based On Image Patch Propagation

This paper focuses on the Criminisi algorithm, the two main aspects of the algorithm about priority calculation method and the searching for the best matching patch are improved. Firstly, we improve the searching of the best matching patch by using the gradient information of the template pixels and the similarity measure of the covariance of the image. Because the gradient conveys very important visual information, and the covariance similarity measure is used to measure the similarity of the two variables, so it is more effective to capture the image structure and the change of the contrast in together. The experimental results show that the covariance gradient method is better than the covariance method, projection method and face recognition method. Then, using the surrounding information of the missing area, eliminating the interference loss of information, we improved the calculation method of the priority. In the end, combining the two aspects, we restore image by using block diffusion theory. Simulation results show that the proposed method can not only improve the ability to repair the damaged image, but also can better maintain the texture and structure of the damaged area.